OpenAI and Paradigm Jointly Release 'EVMbench', a Benchmark for Smart Contract Vulnerabilities

OpenAI, known for developing ChatGPT, announced on February 18th, in collaboration with cryptocurrency investment firm Paradigm, the launch of 'EVMbench' on their official website. This benchmark is designed to measure the capabilities of AI agents in detecting, fixing, and exploiting vulnerabilities in smart contracts. EVMbench consists of 120 severe vulnerabilities extracted from 40 audits and evaluates AI agents in three modes. 'Detection mode' assesses the ability to audit code repositories and identify vulnerabilities, 'Fix mode' tests the capability to safely rewrite vulnerable contracts while maintaining their functionality, and 'Exploit...